Detect binary content beyond the first line

[Matei02355]

Summary

• inspect up to the first 1024 already-buffered bytes before splitting out the first line
• preserve the reader's bytes, line boundaries, UTF-16 handling, and streaming behavior
• add regressions for cross-line binary detection, the 1024-byte boundary, single-read/EOF inputs, and the CLI binary header

Fixes #3554.

Root cause

content_inspector checks up to 1024 bytes for a NUL byte, but bat passed only the first line. Random or encrypted data can contain a newline before its first NUL byte, so the shortened sample was classified as UTF-8 and binary bytes were sent to the terminal.

The fix snapshots the already-buffered prefix with BufRead::fill_buf() before read_until splits out the first line. The snapshot is non-consuming, so no bytes are lost or reordered, and it does not add a post-line blocking read. Empty input is handled without requesting a second EOF event.

Compatibility

BOM detection still takes precedence, so UTF-16 input handling is unchanged. ZIP signature detection is also unchanged.

PR #3763 changes the same input initialization path for a distinct issue (#2262: bounding reads for newline-free binary files). If it merges first, this PR will need a small rebase, but the behaviors are complementary.

Validation

• cargo test --all-targets --all-features — clean run: 455 passed, 5 ignored
• cargo clippy --all-targets --all-features -- -D warnings
• cargo fmt --check
• git diff --check

[keith-hall]

Thanks, looks good